Download Risk assessment and screening for potentially invasive

Kolar—Screening
for of
potentially
invasive
fish Research, 2004, Vol. 38: 391–397
New
Zealand Journal
Marine and
Freshwater
0028–8330/04/3803–0391 © The Royal Society of New Zealand 2004
391
Invited review
Risk assessment and screening for potentially invasive fishes
CINDY KOLAR
U.S. Geological Survey
Upper Midwest Environmental Sciences Center
2630 Fanta Reed Road
La Crosse, WI 54603
United States
email: [email protected]
Abstract Preventing the introduction of potentially invasive species is becoming more important
as this worldwide problem continues to grow. The
ability to predict the identity or range of potential
invaders could influence regulatory decisions and
help to optimally allocate resources to deal with
ongoing invasions. One screening tool presented
here, using species life history and environmental
tolerances to identify potential invaders similar to
past invaders, can be used to predict potential invading species. Another screening tool, genetic algorithms, can be used to predict the potential range of
an invading species. Use and further development
of tools such as these, that are quantitative and relatively transparent, would give managers and other
decision makers more information for making better-informed decisions.
Keywords GARP; genetic algorithms; habitat
niche modelling; invasive species; risk assessment;
species characteristics; species screening
INTRODUCTION
Invasive species are a leading threat to global
biodiversity (Sala et al. 2000; Novacek & Cleland
2001) and are economically costly (Pimentel et al.
2000). To deal with this growing threat, an
increasing number of countries and states require
(e.g., New Zealand and Western Australia) or have
legislation pending that would require (e.g., United
States) the screening of plants and/or animals before
importation to identify those that are potentially
invasive. Screening tools used for this purpose need
to be transparent, relatively easy to use, reliable, and
unbiased (i.e., give the same results independent of
screener) to be defensible and fair (Kolar & Lodge
2002). Ideally, they should also be quantitative—an
attribute that would embody several of the characteristics listed above. Here, I will describe two
quantitative and reliable approaches that can be used
as screening tools for invading species. The first
method, based on the life history characteristics and
environmental tolerances of previous invaders in an
ecosystem (using historical fish invasions in the
Laurentian Great Lakes as an example), can be used
to predict whether a species is likely to become
established, spread, or to be perceived as a nuisance.
The second method, based on ecological niche
modelling of potentially invasive species (using
General Algorithm for Rule-set Predictions, or
GARP, as an example), can be used to predict the
potential range and other effects of non-indigenous
species.
SCREENING TOOLS BASED
ON SPECIES CHARACTERISTICS
M03069; Online publication date 3 August 2004
Received 10 October 2003; accepted 6 July 2004
Screening tools to predict invading species based on
species characteristics assume that invading species
somehow differ from species that tend not to invade.
If invaders can be accurately distinguished from noninvaders, it follows that the identity of potentially
invading species can be predicted. It has been
hypothesised that characteristics such as early
392
New Zealand Journal of Marine and Freshwater Research, 2004, Vol. 38
maturity, high fecundity, asexual reproduction,
having general habitat and diet requirements, and
existing well with humans are associated with successful invading species (see more complete listing
in Lodge 1993). One synthesis of studies examining
characteristics of invading species suggested,
however, that although some patterns were present,
characteristics associated with invading species were
not consistent across locations (Williamson 1996).
In a 2001 literature review of quantitative studies that
limited potential confounding in the data by
examining only one ecosystem, taxon, and stage in
the invasion process at a time (e.g., either
“establishment” or “invasiveness”), Kolar & Lodge
(2001) found that some characteristics of invading
species differed from non-invading species across
studies. They found, for instance, that invasive plants
are more likely to have a history of invasion and to
reproduce vegetatively than non-invasive plants in
particular ecosystems. In addition, some characteristics, such as having a broad diet and diverse
climatic tolerances, previously thought to be
associated with invaders were consistently not so
associated (Kolar & Lodge 2001). This review
located only studies conducted on plants and birds,
however, thereby identifying the need for similar
studies to be conducted for other taxa. Because
invading and non-invading species differed in
respect to some species characteristics, results from
this study indicate some promise using species
characteristics to identify potential invaders.
Until recently, few quantitative species screening
models had been developed for aquatic ecosystems.
Rather, the method most often used for aquatic
species has been deductive risk assessment. This
method, based on making inferences about a risk
based on what is already known or has occurred,
begins with a review of the literature to identify
species with a history of invasion and then uses
selection criteria to identify those with a higher
potential of becoming invasive at a given location
(Hayes & Sliwa 2003). Although the method is not
based strictly on species characteristics, species
attributes can be included in the specific selection
criteria. Deductive risk assessment has been used to
identify potential invaders to the Laurentian Great
Lakes (Ricciardi & Rasmussen 1998), potentially
invasive aquatic weeds in New Zealand (Champion
& Clayton 2000), and potential marine invaders in
Australia (Hayes & Sliwa 2003), among other
applications. A weakness in using the deductive risk
assessment approach is the assumption that all
potential invaders have already invaded elsewhere.
Although this is often true, high impact invading
species in an ecosystem sometimes have no history
of previous invasion (e.g., sea lamprey Petromyzon
marinus in the Laurentian Great Lakes).
Screening approaches based on species characteristics to date have been dominated by efforts for
terrestrial plants. Two of the more widely used weed
risk assessments, the Weed Risk Assessment of
Australia (Groves et al. 2001) and the Weed-Initiated
Pest Risk Assessment guidelines developed by the
Animal and Plant Health Inspection Service of the
U.S. Department of Agriculture (2002), are based on
qualitative or semi-quantitative categorisations of
diagnostic characteristics. Two quantitative
approaches of screening using species characteristics
for plants include Reichard & Hamilton (1997) to
predict invasive woody plants in North America, and
Daehler et al. (2004) to predict invasive plants in
Hawaii. Below is an example of a quantitative risk
assessment constructed using species life history
characteristics and environmental tolerances in one
ecosystem (the Laurentian Great Lakes), for one
taxon (fishes), and for several stages of the invasion
sequence (establishment, spread, and impact),
examined independently.
Historical fish invasions
in the Laurentian Great Lakes as an example
Kolar & Lodge (2002) used the historical establishment of non-indigenous fishes in the Laurentian
Great Lakes to determine if successfully introduced
fishes differed predictably from fishes that were
introduced but did not become established. They
developed statistical models that differentiated fishes
introduced successfully (n = 24 species) from those
that failed to become established in the Great Lakes
(n = 21 species). Data on 26 variables (including 14
life history characteristics, five environmental
tolerances, six aspects of invasion history, and
degree of human use) were obtained for each species.
Statistical models developed from those data identified four variables that accurately discriminated
successful invaders from unsuccessful invaders with
87–96% accuracy using discriminant analysis and
categorical and regression tree analysis (see Kolar
& Lodge 2002). Discriminant functions were also
developed to distinguish quickly-spreading from
slowly-spreading non-indigenous fishes, and those
fishes perceived as a nuisance from those perceived
as a non-nuisance (see Kolar & Lodge 2002).
Kolar & Lodge (2002) used models based on
historical fish invasions in the Great Lakes and life
history information from Eurasian fishes to identify
Kolar—Screening for potentially invasive fish
18 fishes from the Ponto-Caspian Basin, a donor
region for introduced species in the Great Lakes,
with a high risk of becoming established in the Great
Lakes if introduced. Further identified were 5 of the
18 that were also predicted to spread quickly and to
be perceived as a nuisance in the Great Lakes if they
were introduced (see Kolar & Lodge 2002).
Although the species characteristics found to be
associated with invading fishes and the predicted
high-risk species are specific to the Great Lakes, a
major advantage to using this modelling approach
as a screening tool is its applicability to a variety of
ecosystems, taxa, and stages of the invasion process.
Other advantages are that the models are not difficult
to construct using presently available software, and,
although substantially more species characteristic
data are needed to construct the models, only limited
data are required to use them as screening tools. A
disadvantage of this modelling approach is the
limited availability of data on failed introductions,
although these data are probably more available than
may be perceived (e.g., from agency stocking
records, grey literature). Other disadvantages include
data-intensive model development and geographic
limitations to model predictions to the ecosystem for
which the model was developed. Also, the reliability
of these models (and all other models) depends not
only on the accuracy of the prediction, but also the
frequency with which the event occurs at all (i.e., the
“base rate”). Models predicting rare events can be
susceptible to substantial base rate error (Smith et
al. 1999). The effect of base rate on the models in
this example are minimal (see Kolar & Lodge 2002).
SCREENING TOOLS BASED ON
ECOLOGICAL NICHE REQUIREMENTS
Another quantitative method to screen for potentially
invasive species is based on the concept of the
ecological niche, which refers to the “set of
tolerances and limits in multidimensional space that
define where a species is potentially able to maintain
populations” (based on Grinell 1904, 1917; quoted
from Peterson & Vieglais 2001) and includes
ecological tolerances and other factors limiting
distribution. Ecological niche models are developed
to explain geographic distributions and typically
include parameters such as temperature, precipitation, aspect, elevation, and vegetation (Peterson &
Vieglais 2001). Thus the goal of ecological niche
modelling is to describe the fundamental niche of a
species as completely as possible using geographical
393
and ecological data. Once an ecological niche model
is developed for a species, the model can then be
used as a screening tool to predict invasive potential
by projecting the niche parameters outside of the
native range where these requirements can be met.
Areas providing necessary niche requirements would
be expected to support the species if it were
introduced. Thus, ecological niche models and
resulting potential distribution maps can be used to
evaluate the invasive potential of species not yet
found in a given ecosystem.
General Algorithm for Rule-set
Predictions (GARP) as an example
General Algorithm for Rule-set Predictions (GARP)
is a general machine-learning algorithm that has been
used to predict species distributions from geographical and ecological data (Peterson & Vieglais 2001;
Drake & Bossenbroek in press) (available for
download
at
http://www.lifemapper.org/
desktopgarp). Using species presence or absence
data and spatially explicit environmental niche
parameters (such as precipitation, elevation, aspect,
and temperature, available on a global scale at a
fairly fine resolution), GARP searches iteratively for
non-random correlations between species presence/
absence and the niche parameter values using several
different types of rules (atomic, logistic regression,
bioclimatic envelopes, and negated bioclimatic
envelopes) (Stockwell & Noble 1992; Stockwell
1999; Stockwell & Peters 1999). The resulting
ecological niche model can then be tested for
accuracy using point location data from the native
range. Predictions of the potential range of a species
can then be made by projecting the ecological niche
model to identify areas where the species may
become established if introduced (see manual by
Payne & Stockwell 2002, for instructions on use of
GARP).
GARP has been used for a variety of applications,
including modelling spatial patterns of rare species
for conservation purposes (Ortega-Huerta &
Peterson 2004), examining realised and potential
effects of global climate change on species’
distributions (Peterson 2003a,b; Thomas et al. 2004),
investigating evolution and speciation (Peterson et
al. 1999; Barber et al. 2004; Peterson 2004),
predicting the potential ranges of invading species
(garlic mustard and Russian olive, Peterson &
Robins 2003; Asian longhorned beetles, Peterson et
al. 2004; and mosquitoes, Levine et al. 2004), and
examining interaction between range of invasive
species and conservation (barred and spotted owls,
394
New Zealand Journal of Marine and Freshwater Research, 2004, Vol. 38
Peterson et al. 2003a) and bird migration (Peterson
et al. 2003b). GARP is a powerful analytical tool that
can be used for any taxon and allows predictions to
be made up to a global scale at a fairly fine
resolution. Drake & Bossenbroek (in press) state that
GARP is relatively reliable and well validated
compared with similar algorithms for predicting
species occurrences. They cite the benefits of the
algorithm as including the need for as few as five
factors (Peterson & Cohoon 1999), being insensitive
to the lack of absence data (Stockwell & Peterson
2002a), converging using only c. 50 observations
(Stockwell & Peterson 2002b), and use of a best
subset technique to improve model selection and
forecasting (Anderson et al. 2003).
There are several disadvantages to using this
approach, however. Ecological niche modelling
predicts potential range of the non-indigenous
species, but cannot provide information about its
impact. Also, ecological niche modelling assumes
that niche parameters for which geospatial coverages
available are those that limit the distribution of
species, and that the realised niche of a species in
its native range is similar to the fundamental niche
of the species. Ecological niche modelling also
requires extensive knowledge of the native range of
the species and the availability of geospatial and
ecological data sets for the desired variables.
Although the availability of geospatial data sets
continues to improve, there is a need to develop and
make available more coverages to continue to
improve the utility and power of algorithms such as
GARP for predicting potential ranges and other
aspects of the ecology of invasive species.
Drake & Bossenbroek (in press) overcame the
lack of geological geospatial coverages for the native
range of zebra mussels in Europe and Asia and were
able to use GARP to predict the potential range of
zebra mussels in North America based on their
current distribution in North America (n = 2416
observations). They were able to do so by assuming
that zebra mussels have had access to all regions of
the United States, based on their slow rate of
expansion since 1993. Although this method requires
more data on the invaded range of the species, Drake
& Bossenbroek (in press) were able to show that
virtually the same range prediction would be made
with half of the distributional observations. To date,
applications of GARP have been limited primarily
to terrestrial ecosystems. Besides Drake &
Bossenbroek (in press), the only other published
account of using GARP for invasive aquatic species
that I located was the modelling of the potential
distribution of Hydrilla in Peterson et al. (2003a).
There is ongoing work in the United States, however,
using habitat niche modelling to predict the potential
ranges of over 50 freshwater fishes (J. Williams, U.S.
Geological Survey pers. comm.). The limited use of
GARP and similar algorithms to understand and
predict aspects of the invasion of aquatic species is
probably caused by the lack of geospatial data for
these systems. Currently, climatic and geological
data must often be used as surrogates for variables
in the aquatic environment.
USE OF SPECIES CHARACTERISTICS
AND ALGORITHMS TO PREDICT
POTENTIAL INVADING SPECIES
Applications of both of the species screening
methods described herein are dominated to date by
species invasions in terrestrial ecosystems. One
explanation for this discrepancy may be based on
the differential histories of management and control
of invasive species in aquatic and terrestrial habitats.
The desire to protect and enhance crop and pasture
productivity has led to a long history of weed and
crop pest control. In the United States, for example,
the Western Society of Weed Science was formed
in 1938, followed by the Southern Weed Science
Society in 1947. To date, there is no professional
society in the United States focused on aquatic
invasive species (although the International
Conference on Aquatic Invasive Species which held
its 13th annual conference in 2004 is a venue
dedicated to this topic). The direct economic benefit
of protection of crop and livestock forage, estimated
to be US$33 billion annually (Pimentel et al. 2000),
has driven the development of prevention and
control methodologies. Sociological and political
will exists in the United States for continued research and development for the control of terrestrial
weeds. Aquatic habitats are less visible and economic benefits derived from these ecosystems are
often more indirect and for non-market ecosystem
services. In addition, sociological and political will
has diminished in the United States for the
widespread use of chemical or biological control in
aquatic ecosystems. Thus, pressure for making
progress in prevention, control, and management of
invasive species has been greater for terrestrial than
for aquatic habitats. Another explanation for greater
use of species screening systems in terrestrial versus
aquatic ecosystems may be the availability of
relevant data: whether regarding species
Kolar—Screening for potentially invasive fish
characteristics or geospatial in nature, data for
terrestrial ecosystems are often more available than
for corresponding data in aquatic ecosystems.
There are two patterns emerging in the literature
that should be mentioned in regard to their relevancy
to species screening methodologies: the correlation
between propagule pressure and establishment and
biotic resistance as a determinant of invasion
success. Several studies have shown that the likelihood of successful introduction increases with
propagule pressure (or the number of individuals
introduced and the number of introduction events)
(Kolar & Lodge 2001; Duncan et al. 2003; Rouget
& Richardson 2003). Differences in propagule
pressure would not directly affect the predictions
made from species screening models as presented
here. The higher the propagule pressure, the more
likely it would be that an introduced species would
be introduced into a habitat amenable to survival
given the potential range identified through
modelling (or the more likely the species would have
suitable characteristics).
VALUE OF PREDICTION
Assuming that potentially invading species can be
identified, either by the methods described herein or
by others, what is the benefit of doing so? First,
identifying potentially invasive species before their
importation and introduction provides the opportunity to block entry for those high-risk species by
way of legislation and regulation. Second, for those
non-indigenous species already established,
particularly those in the early stages of invasion,
identifying those that pose the greatest risk allows
for the timely allocation of monetary and personnel
resources, and the development of species-specific
management strategies for eradication, containment,
or control.
Neither of the approaches described here to
predict invading species are perfect; errors would
inevitably be made if management decisions were
based solely upon any single type of model predictions. Given this reality, what is the value of using
modelling techniques to predict invading species?
Any technique used to screen species for importation
or any method used to ascertain high priority species
for focused management will be imperfect. Mistakes
will be made whether decisions are based wholly on
qualitative or quantitative assessments. Some of the
benefit of using quantitative modelling approaches
to characterise potential invaders are realised in
395
increased transparency, reliability, and tractability,
and decreased opportunity for subjective influence.
The greatest benefit, however, of using quantitative
approaches for identifying potentially invading
species is knowing the probability of error associated
with model predictions. Understanding that a model
has a 5% or 10% chance of inaccurate predictions
provides managers and policy makers with more
information on which to base decisions rather than
a qualitative assessment or indicator without knowledge of probable error. Given that after establishment, few invading species are successfully
eradicated, and that eradication, containment, and
control are economically costly, prudent management and policy decisions should be based on the
predictions and recommendations of all screening
resources—including expert opinion, qualitative and
quantitative predictions, and new screening tools as
they become available.
ACKNOWLEDGMENTS
Work on using species characteristics to predict invading species was supported by the National Sea Grant
Program (643-1532-04 to David M. Lodge and others)
and a Clare Boothe Luce Presidential Fellowship. Support for the writing of this manuscript was provided by
the U.S. Geological Survey, Upper Midwest Environmental Sciences Center. John M. Drake provided advice and
expertise on genetic algorithms.
REFERENCES
Anderson, R. P.; Lew, D.; Peterson, A. T. 2003: Evaluating predictive models of species’ distributions:
criteria for selecting optimal models. Ecological
Modelling 162: 211–232.
Animal and Plant Health Inspection Service, U.S. Department of Agriculture 2002: Weed-initiated pest
risk assessment guidelines for qualitative assessments, version 5.2. Plant Protection and Quarantine, Riverdale, Maryland.
Barber, B. R.; Martinez-Gomez, J. E.; Peterson, A. T.
2004: Systematic position of the Socorro mockingbird Mimodes graysoniii. Journal of Avian
Biology 35: 195–198.
Champion, P. D.; Clayton, J. S. 2000: Border control for
potential aquatic weeds—Stage 1 weed risk model.
Science for Conservation 141. Wellington, New
Zealand, Department of Conservation.
Daehler, C. C.; Denslow, J. S.; Ansari, S.; Kuo, H. C.
2004: A risk-assessment system for screening out
invasive pests plants from Hawaii and other Pacific Islands. Conservation Biology 18: 360–368.
396
New Zealand Journal of Marine and Freshwater Research, 2004, Vol. 38
Drake, J. M.; Bossenbroek, J. M. 2004: The potential
distribution of zebra mussels (Dreissena
polymorpha) in the U.S.A. BioScience, in press.
Duncan, R.P.; Blackburn, T.M.; Sol, D. 2003: The ecology of bird introductions. Annual Review of Ecology, Evolution, and Systematics 34: 71–98.
Grinnell, J. 1904: The origin and distribution of the chestnut-backed chickadee. Auk 21: 364–382.
Grinnell, J. 1917: Field tests of theories concerning distributional control. American Naturalist 51: 115–
128.
Groves, R. H.; Panetta, F. D.; Virtue, J. G. 2001: Weed
risk assessment. Collingwood, Victoria, Australia,
CSIRO Publishing.
Hayes, K. R.; Sliwa, C. 2003: Identifying potential marine pests—a deductive approach applied to Australia. Marine Pollution Bulletin 46: 91–98.
Kolar, C. S.; Lodge, D. M. 2001: Progress in invasion
biology: predicting invaders. Trends in Ecology
and Evolution 19: 199–204.
Kolar, C. S.; Lodge, D. M. 2002: Ecological predictions
and risk assessment for alien species. Science
298: 1233–1236.
Levine, R. S.; Peterson, A. T.; Benedict, M. Q. 2004:
Geographic and ecologic distributions of the
Anopheles gambiae complex predicted using a
genetic algorithm. American Journal of Tropical
Medicine and Hygiene 70: 105–109.
Lodge, D. M. 1993: Biological invasions: lessons for
ecology. Trends in Ecology and Evolution 8: 133–
137.
Novacek, M. J.; Cleland, E. E. 2001: The current
biodiversity extinction event: scenarios for mitigation and recovery. Proceedings of the National
Academy of Sciences 98: 5466–5470.
Ortega-Huerta, M. A.; Peterson, A. T. 2004: Modelling
spatial patterns of biodiversity for conservation
prioritization in north-eastern Mexico. Diversity
and Distributions 10: 39–54.
Payne, K.; Stockwell, D. R. B. 2002: GARP modelling
system user’s guide and technical reference. San
Diego, United States, San Diego Supercomputing
Center. (Online at: http://biodi.sdsc.edu/Doc/
GARP/Manual/manual.html)
Peterson, A. T. 2003a: Subtle recent distributional shifts
in great plains bird species. Southwest Naturalist
48: 289–292.
Peterson, A. T. 2003b: Projected climate change effects
on Rocky Mountain and Great Plains birds: generalities of biodiversity consequences. Global
Change Biology 9: 647–655.
Peterson, A. T. 2004: The speciation and biogeography
of birds. Science 304: 1249–1249.
Peterson, A. T.; Cohoon, K. P. 1999: Sensitivity of distributional prediction algorithms to geographic data
completeness. Ecological Modelling 117: 159–
164.
Peterson, A. T.; Robins, C. R. 2003: Using ecologicalniche modelling to predict barred owl invasions
with implications for spotted owl conservation.
Conservation Biology 17: 1161–1165.
Peterson, A. T.; Vieglais, D. A. 2001: Predicting species
invasions using ecological niche modelling: new
approaches from bioinformatics attack a pressing
problem. BioScience 51: 363–371.
Peterson, A. T.; Soberon, J.; Sanchez-Cordero, V. 1999:
Conservatism of ecological niches in evolutionary time. Science 285: 1265–1267.
Peterson, A. T.; Sánchez-Cordero, V.; Soberón, J.; Bartley,
J.; Buddemeier, R. H.; Navarro-Siguenza, A. G.
2001: Effects of global climate change on geographic distributions of Mexical Cracidae. Ecological Modelling 144: 21–30.
Peterson, A. T.; Papes, M.; Kluza, D. A. 2003a: Predicting the potential invasive distributions of four
alien plant species in North America. Weed Science 51: 863–868.
Peterson, A. T.; Vieglais, D. A.; Andresean, J. K. 2003b:
Migratory birds modeled as critical transport
agents for West Nile Virus in North America.
Vector-Borne and Zoonotic Diseases 3: 27–37.
Peterson, A. T.; Scachetti-Pereira, R.; Hargrove, W. W.
2004: Potential geographic distribution of
Anoplophora glabripennis (Coleoptera:
Cerambycidae) in North America. American Midland Naturalist 151: 170–178.
Pimentel, D.; Lach, L.; Zuniga, R.; Morrison, D. 2000:
Environmental and economic costs of nonindigenous species in the United States.
BioScience 50: 53–65.
Reichard, S. H.; Hamilton, C. W. 1997: Predicting invasions of woody plants introduced into North
America. Conservation Biology 11: 193–203.
Ricciardi, A.; Rasmussen, J. B. 1998: Predicting the identity and impact of future biological invaders: a
priority for aquatic resources management. Canadian Journal of Fisheries and Aquatic Sciences 55: 1759–1965.
Rouget, M.; Richardson, D. M. 2003: Inferring process
from pattern in plant invasions: a semi-mechanistic model incorporating propagule pressure and
environmental factors. American Naturalist 162:
713–724.
Kolar—Screening for potentially invasive fish
Sala, O. E.; Chapin, F. S. III; Armesto, J. J.; Berlow, R.;
Bloomfield, J.; Dirzo, R.; Huber-Sanwald, E.;
Huenneke, L. F.; Jackson, R. B.; Kinzig, A.;
Leemans, R.; Lodge, D.; Mooney, H. A.;
Oesterheld, M.; Poff, N. L.; Sykes, M. T.; Walker,
B. H.; Walker, M.; Wall, D. H. 2000: Global
biodiversity scenarios for the year 2100. Science
287: 1770–1774.
Smith, C. S.; Lonsdale, W. M.; Fortune, J. 1999: When to
ignore advice: invasion predictions and decision
theory. Biological Invasions 1: 89–96.
Stockwell, D. R. B. 1999: Genetic algorithms II. In:
Fielding, A. H. ed. Machine learning methods for
ecological applications. Boston, Kluwer Academic
Publishers. Pp. 123–144.
Stockwell, D. R. B.; Noble, I. R. 1992: Induction of sets
of rules from animal distribution data: a robust
and informative method of data analysis. Mathematical Computer Simulations 32: 249–254.
Stockwell, D. R. B.; Peters, D. 1999: The GARP modelling system: problems and solutions to automated
spatial prediction. International Journal of Geographical Information Science 13: 143–158.
397
Stockwell, D. R. B.; Peterson, A. T. 2002a: Controlling
bias in biodiversity data. In: Scott, J. M.; Heglund,
P. J.; Morrison, M. L.; Haufler, J. B.; Raphael, M.
G.; Wall, W. A.; Samson, F. B. ed. Predicting
species occurrences: issues of accuracy and scale.
Washington, Island Press. Pp. 537–546.
Stockwell, D. R. B.; Peterson, A. T. 2002b: Effects of
sample size on accuracy of species distribution
models. Ecological Modelling 148:1–13.
Thomas, C. D.; Cameron, A.; Green, R. E.; Bakkenes,
M.; Beaumont, L. J.; Collingham, Y. C.; Erasmus,
B. F. N.; de Siqueira, M. F.; Grainger, A.; Hannah,
L.; Hughes, L.; Huntley, B.; van Jaarsveld, A. S.;
Midgley, G. F.; Miles, L.; Ortega-Huerta, M. A.;
Peterson, A. T.; Philips, O. L.; Williams, S. E.
2004: Extinction risk from climate change. Nature 427: 145–148.
Williamson, M. 1996: Biological Invasions. Population
and Community Biology Series 15. London,
Chapman & Hall.